Sign in

iiinova is part of the Atmosphere — an open, public network where you own your identity.

Martin Degeling

@mrtn3000.chaos.social.ap.brid.gy
14 followers 1 following 33 posts

Freelancer working with AI Forensics and Institute for Strategic Dialogue. Platform auditing, TikTok, Privacy [bridged from chaos.social/@mrtn3000 on the fediverse by fed.brid.gy ]

PostsRepliesMedia
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 26/08/2026
Hörtipp: "Doomscroll - Wie TikTok Jugendliche im Wahlkampf beeinflusst" Von Diana Köhler und Johanna Tirnthal ist gestern als DLF feature gesendet worden. Ganz spannend den beiden Jugendlichen bei ihren TikTok-Session zuzuhören. Und ich durfte ein bisschen den TikTok-Algo Erklären […]
chaos.social
Original post on chaos.social
002 2
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 24/07/2026
The European commission concludes that TikToks default settings on account visibility are not enough to protect minors.. I hope this is only the example they chose for the PR text because there are several other issues not just related to minors, that TikTok should work on 😬
PRESS RELEASE | Jul 24, 2026 | Brussels | 4 min read
Commission preliminary finds TikTok in breach
of Digital Services Act for failing to ensure safe
accounts for minors
PAGE CONTENTS
Today, the European Commission sent TikTok preliminary findings indicating that TikTok
Top accounts of minors do not meet the safety standards required under the Digital Services
Act (DSA).
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 12/06/2026
They still have half a year to make additional changes, like fixing the unusable search and give additional information on how everything works. So we'll monitor that. While it's great these things are happening, I'm disappointed that TikTok is only doing what is strictly necessary and do not […]
chaos.social
Original post on chaos.social
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 12/06/2026
We already described in February 2024 that the ad library was missing key information and was not compliant with the DSA requirements. tiktok-audit.com/blog/2024/Tik-Tok-… It took till December 2025 and pressure from the European Commission for TikTok to commit to making necessary […]
chaos.social
Original post on chaos.social
100 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 12/06/2026
After more than 2 years the TikTok ad library finally got an upgrade! library.tiktok.com/ads The ad details pages now show crucial information that was missing previously like the video description or the pages the ads linked to. They also show […] [Original post on chaos.social]
Screenshot of an ad on the tiktok ad library. New information is marked in red.
100 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 23/04/2026
Expected something better from this... openai.com/index/introducing-openai…
Screenshot of a terminal window testing the openai privacy filter:
Input: » opf "John Smith is the axolotl rodeo champion of 2026" --device cpu
Output: <PRIVATE_PERSON> is the axolotl rodeo champion of 2026


Input: » opf "Martin Degeling is the axolotl rodeo champion of 2026" --device cpu
Output: Martin Degeling is the axolotl rodeo champion of 2026
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 02/04/2026
In December 2025, TikTok made a series of "commitments" to the European Commission on ad transparency — with 2 to 12 months to implement fixes we'd already flagged in 2024. I know noticed the first change. TikTok's notoriously broken ad library search now shows […] [Original post on chaos.social]
Screenshot of TikToks ad library and the search field showing "keywords".
020 2
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 08/12/2025
Ok, replication shows it's not clear what's happening there.. *Some* uploads are automatically labeled. But when you download the videos from #TikTok again, they do not contain any C2PA metadata anymore.
Label shown to a sora2 video uploaded to TikTok. it says "Contains AI-generated media".
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 05/12/2025
The European Comission "accepts TikTok's commitments on advertising transparency" thttps://ec.europa.eu/commission/presscorner/detail/en/ip_25_2940 The issue they commit to fix now, are mostly what we already criticized more then 1.5 years ago. Let's hope the […] [Original post on chaos.social]
Commission accepts TikTok's commitments on
advertising transparency under the Digital
Services Act
PAGE CONTENTS
Today, the European Commission secured TikTok's commitment to provide advertising
Top, repositories that ensure full transparency around ads on its services, as required by the
Digital Services Act (DSA).
Quote(s)
i After extensive engagement with the Commission, TikTok has submitted binding
Related topics | mitments that address all the concerns raised by the Commission in its investigation
Print friendly pdf and preliminary findings from May 2025. These commitments, which ensure full
transparency, include the following:
Contacts for media

« TikTok will provide the full content of the advertisement as it appears in users' feeds,
including the URLs in the link provided in the ad.

« TikTok will update its repository more quickly, ensuring information is available within
a maximum of 24 hours.

« TikTok will provide the targeting criteria selected by advertisers, along with
aggregated user data (including gender, age group and Member State where the
users who were reached are located), enabling researchers to investigate how ads
are targeted and delivered

« TikTok will introduce additional search options and filters, allowing users to find
advertisements more easily.
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 03/12/2025
In their latest risk assessment report (www.tiktok.com/transparency/en/dsa-…) #TikTok claims that AI content with the #C2PA label that is uploaded is automatically labeled. I tested this yesterday with one video and it seems this is not true […] [Original post on chaos.social]
Quote from TikToks latest Risk Assessment report:
TikTok also works with industry groups and civil society to detect undisclosed AIGC. TikTok participates in the Coalition for Content Provenance and Authenticity (‘C2PA”), an open technical standard and content provenance solution that can provide information in a piece of content's metadata about its origins and whether AIGC models were used to create or edit it. This technology helps TikTok to detect AIGC at scale when content with C2PA metadata is uploaded to the Platform. TikTok was the first video-sharing platform to launch
the ability to read Content Credentials from C2PA.

From Q3 2024 to Q1 2025, 4,781,019 videos on TikTok were auto-labelled with the AIGC tag of “Al-generated” inthe EU.
101 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 03/12/2025
We also spent some time trying to understand how these accounts monetize. TikTok is not very transparent about the monetization on the video level, but when simply looking at the minimal thresholds only 13% of the videos where even eligible for monetization. I […] [Original post on chaos.social]
table showing the costs per second generating videos on current models.
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 03/12/2025
Comments below AI Slop: I was quite surprised that even under the most fake-news-like entries the AI-aspect was rarely discussed. In times when even real news are often criticized as fake only 5% mention AI [ caveat: we only parsed the top 50 comments and replies]
Plot showing the Distribution of Videos with at least one comment mentioning Al by Account Topic
Without any comment mentioning Al Videos with at least one comment mentioning Al
100 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 03/12/2025
The Guardian and Spiegel covered our main results focussing on the prevlance of * Anti-migrant fake news www.theguardian.com/technology/2025… * Sexualized AI-generated images of women […]
chaos.social
Original post on chaos.social
100 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 03/12/2025
We just published our follow-up report on AI-generated "slop" on #tiktok. After testing #ai-slop on Instagram & TikTok this summer, showing that TikTok surfaces more AI content in the search results, we Monitored 354 accounts for 4 weeks that amassed 4.5 billion […] [Original post on chaos.social]
Front page of our report titled "Prompt, Upload, Repeat: How Agentic AI Accounts Flood TikTok with Harmful Content"Quick Summary:
Al Forensics uncovered 354 Agentic Al Accounts (AAAs) that
together amassed more than 4.5 billion views and generated over
43,000 posts made almost exclusively with generative Al. These
accounts flood TikTok with sexualized images of women with
childlike features, fake medical advice, and even antisemitic or
racist content, sometimes packaged to look like news reports or
documentaries. Some of the AAAs specializing in posting Al content
prohibited by TikTok stay on the platform for months, reaching
millions of views. AAAs are replacing human content creation with Al
tools and emerging automated pipelines. AAAs also do not label half
of the Al content they post, and we estimate that TikTok is only
actively labeling less than 1.38% of Al content, increasing its
deceptive potential.
101 1
Reposted by Martin Degeling
Gil Durán @gilduran.com · 24/09/2025
1/ A longtime Wired editor just wrote a mush-brained essay about how he totally missed the political rot of Silicon Valley (& still doesn't get it). But in the late 1990s, a Wired journalist warned of a toxic ideology bubbling up from tech. Paulina Borsook has largely been erased. Let's change that
photo of paulina borsook
8773103370 7.3K
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 20/10/2025
[TikTok / LLM Analysis] In the end, each video took around 5 minutes to process. Does not scale to millions very well, but I still prefer it over spending money on some data center. The M2 is also pretty energy efficient compared to separate GPUs (maxed at 50-60 W under full load). And at the […]
chaos.social
Original post on chaos.social
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 20/10/2025
[TikTok / LLM Analysis] We used the music2emo (huggingface.co/spaces/amaai-lab/mus…) model to get a textual description of the mood of the underlying sound and combined screenshots in collages to reduce the input to the vision-capable qwen model.
huggingface.co
Music2emo - a Hugging Face Space by amaai-lab
Upload an audio file to analyze its emotional characteristics. The model predicts mood tags, valence (positivity), and arousal (intensity) scores.
100 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 20/10/2025
[TikTok / LLM Analysis] Our resources are limited and I favor local models over intransparent cloud services. So I build a pipeline around Ollama using a Mac with 32GB RAM and an M2 processor to first process the different channels of the posts […] [Original post on chaos.social]
Flow chart of the data analysis around using ollama with qwen2.5-vl and other tools like whisper and ffmpeg to analyze TikTok posts.
101 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 20/10/2025
[TikTok / LLM Analysis] Last week at AI Forensics we published a report on what #TikTok|s For You feed looked like if you where interested in a global news event like the Nato Summit end of June. We show how the feed favors engagement baiting and conflict […] [Original post on chaos.social]
Screenshof of the front page of an AI Forensics report titled "From FYP to WW3". Subtitell "Analyzing TikToks's Influence in Polarizing Poltical Discourse During the 2025 NATO Summit"
102 2
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 15/09/2025
@boesch.bsky.social die ganze Studie beweist ja eigenltich genau das, weil sie Daten von vor 12 Jahren benutzt haben. Es ist nicht mehr alles auf den original webservern, aber anderswo archiviert.
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 12/09/2025
[Mention of Kirk incident on TikTok] At the same time Instagram is doing way less to take videos offline that, I guess, you would not stay uncensored on any other media. If you search for kirk the first video with now more than 19M views still shows the original content but added a short […]
chaos.social
Original post on chaos.social
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 12/09/2025
[Mention of Kirk incident on TikTok] I monitored some videos that show the raw footage of the shooting yesterday on #TikTok an provided data to this article: www.wired.com/story/charlie-kirk-sh… It's worth noting that as of today TikTok (or the creators) […]
chaos.social
Original post on chaos.social
100 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 11/09/2025
[Mention of Kirk incident on TikTok] TikTok seems to be largely failing to prevent the spread of videos from the shooting for Charlie Kirk. One video has more than 13 Million views already. People see this in their feed and seem to struggle with it. These are […] [Original post on chaos.social]
Comments on TikTok
"wait | wasn't expecting to see the actual footage today"
Reply:
"me either. | thought it was ai "
100 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 22/08/2025
[Problems with YouTubes Search API] I'm mostly complaining about TikToks API, but it seems YouTube is not much better. It's hard to use for replicable research results, researcher found: arxiv.org/pdf/2506.11727
A screenshot of the title of a PDF. It Reads:

Forgetful by Design? A Critical Audit of
YouTube’s Search API for Academic Research
Bernhard Rieder, Adridn Padilla, and Oscar Coromina
010 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 11/07/2025
[TikTok Research API - No Update] A month ago we showed how many videos are missing from the #TikTok Research API, rendering it useless for many purposes. aiforensics.org/work/tk-api We set up a dashboard to monitor improvements. If you didn't check […] [Original post on chaos.social]
Screenshot from the Dashboard monitoring the TikTok Research API. It shows that a video from TikToks CEO was still not available after a month.
002 2
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 03/07/2025
[TikTok Account Dataset published for BTW2025] We've just published a list of 2344 #TikTok accounts from political parties and candidates: osf.io/6gn8x The list is a collaborative effort of multiple researchers including University of Potsdam and IDZ Jena as well as independent […]
chaos.social
Original post on chaos.social
003 3
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 25/06/2025
[TikTok Ad Library minor fix] TikToks Ad Library has many problems which the European Commission is now finally investigating. And it seems to have an impact. Somebody spent 5 minutes to fix one of the annoying usability issues. The default date of the search […] [Original post on chaos.social]
Screenshot of the default filter of the ad library showing the default date range 25.06.2024 to 25.06.2025
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 13/06/2025
[TikTok & Data Access / Tech Policy] I've been working with #TikTok data for some time now. But what never helped, was trying to use the data access methods they are offering. Together with other researchers we've now summarized our complaints […]
chaos.social
Original post on chaos.social
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 10/06/2025
At AI Forensics we monitored the polish election and found a) #TikTok is still bad at labeling election related content, and b) there were several election fraud conspiracy theories that were not labeled and were seen by millions of people: aiforensics.org/work/tiktok-polish-…
aiforensics.org
Tiktok's Polish Elections Labels: Only Sometimes, And Only For Some
Our investigation reveals TikTok's systematic failures in applying election information labels during Poland's 2025 Presidential Election. The platform excluded 20+ million Polish diaspora users from seeing crucial election labels, while allowing election fraud allegations and AI-generated disinformation to spread unlabeled.
002 2
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 04/06/2025
@boesch.bsky.social We're working on a report that will be out soon on what is wrong with the API data!
020 2
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 20/05/2025
I gave some comments for why the ad library is broken for this article: www.techpolicy.press/experts-echo-e…
techpolicy.press
Experts Echo EU Concerns Over TikTok’s Ad Transparency Gaps | TechPolicy.Press
The European Commission’s preliminary findings echo longstanding concerns from researchers about ad transparency, writes Ramsha Jahangir.
010 1
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 15/05/2025
In February 2024 our team at then SNV published a report on the #TikTok Ad Library and it's short comings: tiktok-audit.com/blog/2024/Tik-Tok-… Today the European Commission announced that their investigations, which included an interview with us) also conclude that the ad library […]
chaos.social
Original post on chaos.social
122 2
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 02/04/2025
To their defense. It got better. After multiple tries the VCE ran a query for 100 videos in just over 80 seconds. Imagine the actual app being that slow 🤷
000 0
Martin Degeling @mrtn3000.chaos.social.ap.brid.gy · 02/04/2025
Trying to work with TikToks Virtual Computing Environment (VCE). Requested data on 10(!) Videos.. 20 minutes passed. still waiting.
screenshot of the TikTok VCE listing "check query task status: Processing" a 100 times
101 1